Abstract
BACKGROUND: Abdominal perfusion pressure (APP) is a salient feature in the design of a prognostic model for patients with intra-abdominal hypertension (IAH). However, incomplete data significantly limits the size of the beneficiary patient population in clinical practice. Using advanced artificial intelligence methods, the authors developed a robust mortality prediction model with APP from incomplete data. METHODS: The authors retrospectively evaluated the patients with IAH from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database. Incomplete data were filled in using generative adversarial imputation nets (GAIN). Lastly, demographic, clinical, and laboratory findings were combined to build a 7-day mortality prediction model. RESULTS: The authors included 1354 patients in this study, of which 63 features were extracted. Data imputation with GAIN achieved the best performance. Patients with an APP <60 mmHg had significantly higher all-cause mortality within 7-90 days. The difference remained significant in long-term survival even after propensity score matching (PSM) eliminated other mortality risks between groups. Lastly, the built machine learning model for 7-day modality prediction achieved the best results with an AUC of 0.80 in patients with confirmed IAH outperforming the other four traditional clinical scoring systems. CONCLUSIONS: APP reduction is an important survival predictor affecting the survival prognosis of patients with IAH. The authors constructed a robust model to predict the 7-day mortality probability of patients with IAH, which is superior to the commonly used clinical scoring systems.